According to the theory of reasoned action of Fishbein&Ajzen (1975), beliefs, attitudes, intentions and behaviour form a causal chain, so that beliefs lead to attitudes, and attitudes in turn lead to intentions and so behaviour. Despite the undoubted influence of the model, a paper by Towriss (1984) has pointed to a discrepancy between what the model says and the way in which it is usually tested: while the model is aboutindividuals'beliefs, subjects are normally asked aboutmodalbelief statements supplied by the experimenter. To examine whether the difference is critical, the present experiment was designed to test both approaches, by asking students about drinking milk. Forty subjects were supplied with belief statements (Fishbein&Ajzen condition) and 39 were asked to give their own (Towriss condition). The relationships between beliefs, attitudes, intentions and behaviour were then examined by correlation and multiple regression, and the Towriss condition was found to produce the stronger predictions for intentions and present behaviour, though not for follow‐up behaviour eight weeks later. The paper concludes with a discussion of the theoretical and empirical implications of the finding